Mobile Proxy Startup Data

Startup Data Sources That Hold Up

Most startup datasets are built substantially on press releases, which is why they drift from reality. A funding announcement is a communications decision: made when it suited the company, phrased to flatter, and frequently rounded. A registry filing is a legal statement made under obligation, recording an event on the date it happened.

Building on the statutory record instead changes what the dataset can support. Share allotments, director appointments and departures, registered address changes, accounts, and charges over assets are all filed because they must be, which makes them boringly reliable in exactly the way an analysis needs.

Where Aggregators Fit

Crunchbase, PitchBook and CB Insights sit a layer above the primary sources: they parse announcements and standardise them into one schema, which is where their real value is — a seed round in Singapore and a Series B in Toronto land in the same table. That standardisation also loses fidelity. Two companies trading under the same name in different states get merged into a single row more often than either provider admits, and a valuation attached to a round is frequently the founder's own claim, published with no primary document behind it. Treat an aggregator as an index into which registries to check, not as the record itself.

Company Registries by Jurisdiction

The defining constraint is that there is no global register. The United Kingdom runs Companies House with unusually open bulk data. The United States incorporates at state level, with federal securities filings covering exempt offerings that reveal private rounds. Most European countries operate a national business register with widely varying openness, and OpenCorporates aggregates many of them without replacing any of them.

Several of those national services degrade or restrict access from foreign address ranges, and some render search behaviour differently depending on where the visitor appears to be. Reaching each register from inside its own jurisdiction removes an entire class of intermittent failure that is otherwise extremely hard to attribute — you simply get fewer results, with no indication that anything was withheld.

What Filing Regimes Actually Require

The filing obligation varies enough between jurisdictions that a dataset built to one country's rhythm will silently under-collect in another. A UK small company can file abbreviated accounts that omit a profit-and-loss statement entirely, so revenue is often the one field the registry will never give you. US companies incorporate at state level — Delaware alone hosts a majority of venture-backed startups — and the states themselves publish almost nothing beyond formation date and registered agent; the federal Form D filed with the SEC for an exempt offering is usually the only public record that a private round happened at all, and it reports a raise ceiling rather than the amount actually collected. Germany's Handelsregister and the accompanying Bundesanzeiger split the record in two: corporate facts in one system, annual accounts in the other, under separate access rules. None of this is a reason to skip a jurisdiction — it is a reason to record what a filing regime structurally cannot tell you, so an empty field reads as absence-by-design rather than as a gap in collection.

Hiring as a Growth Signal

Job postings are the earliest public indicator of intent that a company produces. A firm opening roles in a new country is expanding there months before any announcement. A shift in the balance between engineering, sales and support signals a change of stage. A hiring freeze appears as postings quietly disappearing, long before anybody says anything.

Postings are also timestamped and location-bearing by nature, which makes them unusually easy to turn into a series. The trap is that boards and career pages surface roles by the visitor’s region, so a listing aimed at one market may not appear elsewhere. That produces a specific and quite misleading failure: a company looks as though it stopped hiring in a country when in fact you stopped being shown those roles.

Career Pages vs Job Boards

A role usually appears on a company's own career page before it reaches a job board, sometimes by days, because syndication to LinkedIn or Indeed runs through a batch feed rather than in real time. Scraping the career page directly removes that lag and also removes the board's own filtering — some boards deduplicate or suppress postings that resemble reposts, which quietly drops exactly the roles that reopen because a hire fell through. The tradeoff is that every company runs a different careers stack — a hosted Greenhouse or Lever board, a custom page, or nothing machine-readable at all — so a career-page pipeline needs a per-company adapter, where one job-board integration would have covered thousands of companies in a single pass.

company           …
role_title        …
function          engineering | sales | support | ops
location          …
seniority         …
first_seen        2026-05-02   # you derive these
last_seen         2026-08-24
viewed_from       DE           # the exit market that could see this posting

Keeping a Startup Dataset Current

Registry data updates when something is filed, which is irregular and often late — accounts in particular arrive near the deadline rather than near the year end. Hiring data changes daily. Those two rhythms suggest an obvious split: poll postings daily, sweep registries weekly, and treat the filing date rather than the collection date as the event date.

• Key on the registered number — Companies rename constantly. The registration identifier is the only stable key you will get.
• Store the filing date and the collection date — The gap between them is frequently months and is itself informative.
• Reconcile announcements against filings — Where they disagree, the filing is the fact and the difference is a finding.
• Treat officers as personal data — Directors and beneficial owners are named by design, and several jurisdictions limit bulk reuse. Take your own legal advice before aggregating individuals across registers.

Matching the Same Company Across Sources

A registry, a job board and a funding database will rarely agree on how to spell a company's name — trading names change, subsidiaries file under a holding entity, and transliteration alone produces several valid spellings for a company registered outside English-speaking jurisdictions. Matching on name and country catches most of the easy cases and fails silently on the rest, which is worse than an obvious failure because nothing signals that two rows are actually one company. Where a jurisdiction exposes a public identifier — a company number, a VAT ID, a legal entity identifier — key on that instead, and treat name matching as a fallback for the sources that never had one to begin with.

For the neighbouring public-register material see government data, and for the sampling design behind multi-market comparison, market research.

Reach Every National Register

Live PXM2 locations — pick the jurisdictions whose registries and job markets you follow:

Frequently Asked Questions

Related Mobile Proxy Guides

Startup research is register work, which puts it beside the government and nonprofit sources.

Business use cases

Core mobile proxy guides